#!/usr/bin/env bash
set -Eeuo pipefail
shopt -s inherit_errexit nullglob

# -----------------------------------------------------------------------------
# CONFIGURATION
# -----------------------------------------------------------------------------
readonly BACKUP_BASE=/media/extension/backup/ubuntu
readonly MOUNT_POINT=/media/extension
readonly NOW=$(date +'%Y-%m-%d')
readonly THIS_RUN="${BACKUP_BASE}/${NOW}"
readonly LATEST="${BACKUP_BASE}/latest"

# Retention policy
readonly KEEP_DAILY=3    # keep this many daily snapshots
readonly KEEP_WEEKLY=2   # keep this many weekly (Sunday) snapshots

# USB backup drive — ASMedia ASM1153E bridge (174c:55aa)
# Used to disable UAS (which causes resets under load) and spin-down
readonly USB_VENDOR_PRODUCT="174c:55aa"
readonly UAS_QUIRKS_FILE="/etc/modprobe.d/usb-backup-uas-quirk.conf"

# Directories (or files) to back up (without DB volumes)
declare -a SOURCE_ITEMS=(
  "/home/oster/server"
#  "/home/oster/photo-lib"
)

# Rsync exclusions — exclude DB-related dirs that are dumped separately + common junk
declare -a EXCLUDES=(
  --exclude='.cache'
  --exclude='node_modules'
  --exclude='*.tmp'
  --exclude='**/postgresql'
  --exclude='**/mysql'
  --exclude='**/db'
  --exclude='**/database'
)

# Rsync options
RSYNC_OPTS=(
  -aAXHv                # preserve everything, human-friendly, verbose
  --delete              # mirror deletions
  --partial             # keep partials if interrupted
  --stats               # summary
  "${EXCLUDES[@]}"
)

# Link-dest array — populated later; using an array avoids unquoted-variable issues
LINK_DEST=()

# -----------------------------------------------------------------------------
# LOGGING
# -----------------------------------------------------------------------------
log() { echo "[$(date +'%F %T')] $*"; }
die() { log "FATAL: $*"; exit 1; }

# -----------------------------------------------------------------------------
# FUNCTIONS
# -----------------------------------------------------------------------------
check_mount() {
  if ! mountpoint -q "$MOUNT_POINT"; then
    die "Backup mountpoint $MOUNT_POINT is NOT mounted."
  fi
}

# Ensure the UAS quirk is configured for the backup drive.
# UAS causes device resets under sustained write load with the ASM1153E bridge.
# If the config is missing, write it and update initramfs (requires reboot to take effect).
ensure_uas_quirk() {
  local vid="${USB_VENDOR_PRODUCT%%:*}"
  local pid="${USB_VENDOR_PRODUCT##*:}"
  local expected="options usb-storage quirks=${vid}:${pid}:u"

  if grep -qsF "$expected" "$UAS_QUIRKS_FILE" 2>/dev/null; then
    # Check if the drive is still bound to uas (quirk active only after reboot)
    if ls /sys/bus/usb/drivers/uas/ 2>/dev/null | grep -q .; then
      log "WARNING: UAS quirk is configured but drive is still using the uas driver."
      log "         A reboot is required for the quirk to take effect."
      log "         Backup will proceed but may encounter USB resets."
    else
      log "UAS quirk active — drive is using usb-storage driver."
    fi
  else
    log "UAS quirk not found — writing $UAS_QUIRKS_FILE and updating initramfs."
    echo "$expected" > "$UAS_QUIRKS_FILE"
    update-initramfs -u
    log "Initramfs updated. Please reboot for the UAS quirk to take effect."
    log "Continuing backup with current (uas) driver — may encounter resets."
  fi
}

# Disable drive spin-down and APM for the duration of the backup.
# The drive going to standby mid-write causes "Logical unit not ready" errors.
disable_spindown() {
  local dev
  # Resolve the underlying block device from the mount point (strip partition number)
  dev=$(findmnt -n -o SOURCE "$MOUNT_POINT" | sed 's/[0-9]*$//')
  if [[ -b "$dev" ]]; then
    log "Disabling spin-down and APM on $dev for backup duration."
    hdparm -B 255 -S 0 "$dev" > /dev/null \
      && log "  hdparm: spin-down disabled on $dev." \
      || log "WARNING: hdparm failed on $dev — drive may still spin down."
  else
    log "WARNING: Could not determine block device for $MOUNT_POINT (got: $dev) — skipping hdparm."
  fi
}

# Rotation: keep KEEP_DAILY daily and KEEP_WEEKLY weekly snapshots
rotate_old() {
  log "Applying retention policy (${KEEP_DAILY} daily, ${KEEP_WEEKLY} weekly)."

  # Pre-compute the dates to keep so we don't fork date(1) inside every inner loop iteration
  declare -A keep_dates=()
  for i in $(seq 0 $(( KEEP_DAILY - 1 ))); do
    keep_dates["$(date -d "-$i days" +%F)"]=1
  done
  last_sunday=$(date -d "last sunday" +%F)
  for i in $(seq 0 $(( KEEP_WEEKLY - 1 ))); do
    keep_dates["$(date -d "${last_sunday} -$(( i * 7 )) days" +%F)"]=1
  done

  # Collect snapshot dirs matching YYYY-MM-DD using a glob (nullglob makes empty glob safe)
  for dir in "$BACKUP_BASE"/[0-9][0-9][0-9][0-9]-[0-9][0-9]-[0-9][0-9]; do
    [ -d "$dir" ] || continue
    name=$(basename "$dir")

    if [[ -v keep_dates["$name"] ]]; then
      log "  Keeping snapshot $name"
    else
      log "  Deleting expired snapshot $name"
      rm -rf "${BACKUP_BASE:?}/$name"
    fi
  done
}

# -----------------------------------------------------------------------------
# MAIN
# -----------------------------------------------------------------------------
log "=== Starting backup run: $NOW ==="
check_mount
ensure_uas_quirk
disable_spindown

# Fail early if the backup base directory doesn't exist (drive not mounted / path wrong)
[ -d "$BACKUP_BASE" ] || die "Backup base directory $BACKUP_BASE does not exist."

# Must be able to write into the backup base (e.g. run as root via sudo/cron)
[ -w "$BACKUP_BASE" ] || die "No write permission on $BACKUP_BASE — run as root (sudo)."

# Locking — base dir is confirmed to exist, so the lockfile write will succeed
lockfile="${BACKUP_BASE}/.backup.lock"
cleanup() { rm -f "$lockfile"; }
trap cleanup EXIT
if ! ( set -o noclobber; echo "$$" > "$lockfile" ) 2>/dev/null; then
  # Check if the PID in the lockfile is still alive; if not, it's a stale lock
  stale_pid=$(cat "$lockfile" 2>/dev/null || true)
  if [[ -n "$stale_pid" ]] && ! kill -0 "$stale_pid" 2>/dev/null; then
    log "Removing stale lockfile (PID $stale_pid is no longer running)."
    rm -f "$lockfile"
    echo "$$" > "$lockfile"
  else
    die "Another backup is already running (PID ${stale_pid:-unknown}, lockfile: $lockfile)."
  fi
fi

mkdir -p "$THIS_RUN"

# Incremental snapshot?
if [ -L "$LATEST" ] && [ -d "$LATEST" ]; then
  latest_real=$(readlink -f "$LATEST")
  LINK_DEST=("--link-dest=${latest_real}")
  log "Using link-dest for incremental snapshot → $latest_real"
else
  LINK_DEST=()
  log "No previous backup; creating full snapshot."
fi

# Run backups
for src in "${SOURCE_ITEMS[@]}"; do
  base=$(basename "$src")
  dst="${THIS_RUN}/${base}"
  mkdir -p "$dst"
  log "Backing up $src → $dst"
  nice -n 10 ionice -c2 -n7 \
    rsync "${RSYNC_OPTS[@]}" "${LINK_DEST[@]}" \
      "$src"/ "$dst"/
done

# Update latest symlink atomically
ln -sfn "$(basename "$THIS_RUN")" "$BACKUP_BASE/.latest.tmp"
mv -T "$BACKUP_BASE/.latest.tmp" "$LATEST"

log "Snapshots done. Running rotation policy..."
rotate_old

log "=== Backup completed successfully ==="